Pilot Glade on a real project

Adoption guide

Compare one representative local workflow with Salesforce before making Glade a team or CI gate.

Glade does not replace Salesforce. This pilot establishes where a faster local loop is useful and where hosted validation remains required.

1. Choose a representative path

Pick one project and one frequent job: check Apex, run one test class, debug a test, or run changed tests. Avoid starting with the largest suite or a path dominated by hosted services.

Record:

• the Salesforce DX project and package directory
• the exact Salesforce command/result used as the comparison
• the Glade version under evaluation
• whether the path uses live auth, hosted services, or org-only metadata

2. Establish project context

curl -fsSL https://glade.sh/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"
glade version
cd path/to/salesforce-dx-project
test -f glade.yml || glade init --project . --yes
glade config validate --project .
glade doctor --project .

Expected: doctor identifies the project and ends with Ready. before the pilot uses source checks or tests.

3. Run the local equivalent

glade check --project .
glade test --project . --class RefinementServiceTest --json

Use an actual class from the project. Save the command, exit code, diagnostic or test result, runtime, and any local files written.

4. Compare the evidence

Compare result semantics, not only whether both commands passed. Record:

• diagnostics and source locations
• selected tests and pass/fail outcomes
• changed local data or fixtures
• named unsupported behavior
• the matching Salesforce result

Use What Glade runs locally to classify a difference as a supported local path, a named limit, or behavior that requires Salesforce.

5. Try the daily interface

For VS Code, install the bundled extension and open Glade Home:

glade editor doctor vscode
glade editor install vscode --force

Confirm one diagnostic navigation, Test Explorer or CodeLens run, and—if it is part of the chosen path—one debug session or Apex & SOQL scratch buffer.

6. Add a nonblocking CI trial

Use the pinned advisory pilot workflow. Its assessment and test steps use continue-on-error: true; setup failures remain failures, and artifacts upload with if: always(). Full git history is needed for affected-test selection. Do not make this job a required merge gate during evaluation.

Keep the trial advisory until representative local results have been compared with Salesforce. Promote only the paths the team has reviewed.

7. Exercise one deeper workflow

Choose only what the pilot needs:

• AI-assisted Apex for a repeatable agent check/fix/rerun contract
• glade report refactor-proof --project . --since origin/main for change evidence
• glade dev vf --project . --addr 127.0.0.1:8080 for supported local Visualforce preview
• glade dev lwc --project . --open for the LWC Workbench Console
• Plugins when the default public plugin registry serves a needed first-party package

Exact hosted Visualforce, Lightning Experience, live auth, deployment, and production governor behavior remain Salesforce validation work.

8. Decide and report

Useful pilot feedback includes:

• glade version
• full glade doctor --project . output
• the exact command and exit code
• the smallest source, metadata, fixture, or saved debug log that shows a mismatch
• the Salesforce comparison result
• the support-map classification

Open a reproducible GitHub issue for an unexplained supported-path mismatch. Use neutral project labels. Do not include proprietary source, private package names, credentials, customer records, or unredacted support bundles. Vulnerabilities belong in private security reporting, not public issues.

Record a small manual table; no new telemetry is required:

Evaluation	Record
Environment	Neutral project label, OS/architecture, product/plugin versions, source API version
First run	Install/doctor outcome, named tests selected/executed/passed, time to first useful result
First mismatch	Expected/actual behavior, support classification, issue link and owner
Next step	Could the tester tell what to do next? Would they use this loop again, and for what?

Start with 5–10 approved evaluators and report actual n/N outcomes. Try the sample before a focused test in a real project the tester is authorized to use. Fix repeated setup failures and false-success behavior before broader promotion.

Pilot completion

The pilot is complete when the team can name which local paths are useful, which paths remain advisory, which paths still require Salesforce, and what evidence a future CI gate will retain.